Maximizing System Lifetime in Wireless Sensor Networks

swarmtellingMobile - Wireless

Nov 21, 2013 (3 years and 7 months ago)

73 views

Maximizing SystemLifetime in Wireless Sensor Networks
Qunfeng Dong
Department of Computer Sciences
University of Wisconsin
Madison,WI 53706
qunfeng@cs.wisc.edu
ABSTRACT
Maximizing system lifetime in battery-powered wire-
less sensor networks with power aware topology control
protocols and routing protocols has received intensive
research.In the past,this problem has been mostly
studied from the indirect perspective of energy conser-
vation.Although this leads to solutions that help ex-
tend network lifetime,energy conservation is not the
same problem as network lifetime maximization.Some
researchers have formally studied network lifetime max-
imization problems,based on the assumption that en-
ergy is only consumed by packet transmission.How-
ever,it is well known that in many cases energy is sig-
nificantly consumed during overhearing and idle peri-
ods.In this paper,we try to present formal analysis
of a variety of network lifetime maximization problems
in different energy consumption models.In particular,
we identify different energy consumption models,define
a variety of fundamental network lifetime maximization
problems in individual energy consumption models,and
formally analyze their complexity.Polynomial time al-
gorithms are presented for tractable problems,and NP-
hardness proofs are presented for intractable problems.
1.INTRODUCTION
Multi-hop,ad hoc,wireless sensor networks (WSNs)
are considered a promising technology to change our
physical environment and hence our life in this envi-
ronment.WSNs are typically deployed using battery-
powered stationary sensor nodes equipped with sensing,
computing and wireless communicating modules.In a
broad range of potential applications,inexpensive sen-
sors can be embedded into buildings or scattered into
spaces to collect,process,store and send out relevant
information for various civilian or military purposes.
When a data sink (e.g.a base station) is out of reach of
a data source sensor node,they can rely on intermediate
sensor nodes to relay data packets.
A salient feature of battery-powered WSN is its ex-
tremely constrained source of energy supplied by bat-
teries coming with sensor nodes,because sensor nodes
are typically small and thus use tiny batteries.In many
scenarios,it seems infeasible to replace or recharge bat-
teries of sensor nodes.For example,NASA plans to
deploy sensor networks in areas of interest on Mars [1].
Meanwhile,in WSNs,wireless communication is con-
sidered much more energy consuming than sensing and
computing [2].All these factors make it essential to de-
velop efficient routing and topology control protocols to
maintain requested network properties (e.g.connectiv-
ity) for as long a network lifetime as possible.
In the literature,there have been two different ap-
proaches to maximizing network lifetime.One indirect
approach aims to minimize energy consumption,while
the other approach directly aims to maximize network
lifetime.Although the indirect approach can help ex-
tend network lifetime,it does not address precisely the
problem of maximizing network lifetime.Therefore,
some researchers have aimed to directly maximize net-
work lifetime.
• Chang and Tassiulas [3,4] considered the problem
of maximizing the time to the first node failure for
a unicast session,where each data source generates
data for delivery at a fixed rate.
• In [5,6],optimal solutions are presented for maxi-
mizing the time to the first node failure for a static
broadcast tree.
• In the more general multicast paradigm,Das et
al.[7] presented an optimal solution for maximiz-
ing the time to the first node failure for a static
multicast tree.Flor´een et al.[8] investigated the
problem of maximizing the lifetime of a multicast
session over a network of energy constrained nodes,
where the multicast tree can be dynamically ad-
justed to utilize any node with available energy.
While these efforts are based on the energy consumption
model where energy is consumed only when transmit-
ting packets,it is well known that wireless transceivers
consume a significant amount of energy during over-
hearing and idle periods as well [9,10,11].
The contribution of this paper is the formal analysis
of a number of network lifetime maximization problems,
under different energy consumption models.In par-
ticular,we identify representative energy consumption
models,define a variety of fundamental network life-
time maximization problems under these models,and
formally analyze their complexity.Polynomial time al-
gorithms are presented for tractable problems,and NP-
hardness proofs are presented for intractable problems.
Despite significant research in this area,we do not know
of any optimal solutions to these fundamental problems
identified in this paper,and the complexity of these
problems remain unknown.To the best of our knowl-
edge,this paper is the first to present such a formal
analysis.
The rest of the paper is organized as follows.In Sec-
tion 2,we identify representative energy consumption
models and define network lifetime in individual en-
ergy consumption models.In Section 3,various net-
work lifetime maximization problems are defined under
individual energy consumption models.The complex-
ity of these problems is formally analyzed.Finally,we
conclude the paper in Section 4.
2.MODELS AND DEFINITIONS
In most of the past research efforts aiming to extend
network lifetime,energy consumption is completely at-
tributed to packet transmission.Wireless transceivers
are assumed to consume power only when transmitting
packets,and energy is thus consumed on a per packet
basis.This model is simple and neat.In this paper,
we also include this energy consumption model in our
analysis.For simplicity,we refer to this model as the
packet based model.
Despite the prevalence of the packet based model,it
has been well known that energy is also significantly
consumed during overhearing and idle periods [9,10,
11].In particular,wireless transceivers are powered
to receive every incoming packet and decode to decide
if the packet should be accepted,forwarded,or dis-
carded.Although many packets turn out to be sim-
ply discarded,their reception has already consumed
a significant amount of energy.In addition,wireless
transceivers also consume energy during idle periods,
because they have to be powered to detect if there are
packets being transmitted at all.Researchers [9,11]
have shown that in some cases,energy consumption
during overhearing and idle periods can be comparable
to energy consumption due to transmitting/receiving
packets.In many applications,WSNs are presumed
to be densely deployed,and this has two implications.
On one hand,pair-wise distance between sensor nodes
is small,and thus packet transmission between sensor
nodes consumes less energy.On the other hand,each
sensor node covers more sensor nodes in its transmis-
sion range,and thus more energy will be consumed due
to overhearing.
In the extreme case where wireless transceivers stay
idle and no communication happens at all,energy is
completely consumed in the idle state,on a per time
unit basis.We hereby refer to this energy consumption
model as the time based model.In a broad range of ap-
plications where sensor nodes only need sporadic (and
possibly asynchronous) communication,power consump-
tion is dominated by idle time and transceivers consume
almost the same amount of energy.For example,sen-
sor nodes may be configured to send back environment
information once per hour.The time based model fits
well into such scenarios.In such scenarios,to effec-
tively conserve energy and extend network lifetime,it
is no longer adequate to simply optimize transmission
power as has been done by most researchers.Instead,
we need to turn off as many transceivers as much as
possible.When a sensor node’s transceiver is turned
off,it is considered sleeping.In the sleeping state,en-
ergy consumption during overhearing and idle periods
is avoided.Communication is handled by a backbone
composed of nodes that do not sleep,connecting every
pair of nodes in the network.Asleeping node may occa-
sionally wake up to send out packets over the backbone.
That part of the energy consumption can be addressed
by the packet based model.
In cases where communication is relatively frequent,
energy consumption can be divided into two parts.On
one hand,(homogeneous) sensor nodes consume as much
power as each other on a per time unit basis,due to
overhearing and staying idle.On the other hand,they
may consume significantly different amounts of energy
on a per packet basis,due to packet transmission/reception.
We refer to this case as the mixed model.
Various definitions of network lifetime have been pro-
posed for different scenarios.In [12],Blough and Santi
present a discussion on defining network lifetime,and
outline the principle that network lifetime should refer
to the capability of the network to serve its design pur-
pose.In this paper,we define network lifetime for a
number of network lifetime maximization problems ac-
cording to this general principle.For problems in the
packet based model,we define network lifetime as the
number of packets (to be perfectly accurate,the number
of bits) that can be delivered by the network.This def-
inition applies to all routing paradigms including uni-
cast,multicast and broadcast.
In the time based model,the design purpose is to
maintain an always active communication backbone con-
necting every pair of nodes in the network.Accordingly,
we define network lifetime to be the time until no such
backbone can be formed.This definition is also moti-
vated by the following features of WSNs.
• On one hand,sensor nodes are presumed to be
densely deployed and sensor networks are thus highly
redundant.Even if some sensor nodes fail due
to battery depletion,the whole sensor network is
most likely still in good order to serve its purpose.
• On the other hand,wireless communication is con-
sidered the primary cause of energy consumption
in WSNs,especially in many applications where
sensor nodes only need to conduct modest data
collecting and processing.Even if this assump-
tion is not true in some cases,we may reserve a
certain amount of energy for sensing,processing
and sending data,and reserve the rest of avail-
able energy for staying active in the backbone and
relaying packets.Thus,even if some sensor node
has run out of its energy for relaying packets,it can
still collect,process and send out data as usual.As
long as there exists such a backbone,the function-
ality of the whole sensor network remains intact to
serve its design purpose.
3.ANALYSIS
In this section,we formally analyze a variety of net-
work lifetime maximization problems in the time based
model as well as the intensively researched packet based
model.Definitions and complexity analysis of problems
specific to individual models are presented.Note that
the time based model and the packet based model are
both special cases of the mixed model,thus their hard-
ness results trivially apply to the mixed model.
In our network model,stationary sensor nodes are as-
sumed to be equipped with an omnidirectional antenna.
A wireless sensor network is denoted by a weighted di-
rected graph G = (V,A),where V is the set of sensor
nodes and A is the set of directed links.Each node is
labeled with a unique ID i ∈ [1..|V |] and has a maxi-
mum transmission power of P
i
max
.Let P
ij
denote the
minimum transmission power required to maintain a
reasonably good quality link from node i to node j.G
contains link (i,j) (i.e.,the link from node i to node j)
if and only if P
ij
≤ P
i
max
.Initially,each sensor node
i ∈ V has an energy of p
i
.Time is divided into discrete
time slots,denoted by t ≥ 1,t ∈ Z
+
.
3.1 The time based model
In this section,we investigate the problem of max-
imizing network lifetime in the time based model and
prove it to be NP-hard.When proving the NP-hardness
of intractable problems identified in this section,we ac-
tually prove stronger results that the problems remain
NP-hard even if they are restricted to the special case
where all sensor nodes have the same maximum trans-
mission power P
max
and P
ij
= P
ji
for each node pair
(i,j).In this case,a stationary wireless sensor net-
work can be modelled as a weighted undirected graph
G = (V,E),where E is the set of undirected edges and
G contains edge (i,j) if and only if P
ij
≤ P
max
.
To maintain network connectivity,what we need is
a backbone,which is represented as a connected domi-
nating set (CDS) [13].In an undirected graph G(V,E),
a dominating set is defined as a subset S ⊆ V of nodes
such that each node i ∈ V is either in S or adjacent
to some node v ∈ S.A connected dominating set S
is a dominating set such that the subgraph G

= (S ⊆
V,E

⊆ E) induced by S is connected.Here,we shall
prove an even stronger result that the problemof maxi-
mizing network lifetime while preserving connectivity in
undirected graphs remains NP-hard even if we restrict
it to the special case where during each time step,each
node i ∈ V consumes p
i
energy.In this case,each
node has a battery life of one (time slot) and can be
used in exactly one CDS.The problem of maximizing
network lifetime thus becomes the connected domatic
number (CDN) problem,which is defined as follows.
Connected Domatic Number (CDN)
INSTANCE Graph G = (V,E).Positive
integer K.
QUESTION Does G contain at least K
disjoint CDSs?
Theorem 1.Connected domatic number is NP-hard.
Proof.We prove the NP-hardness of CDNby reduc-
ing from the 3-dimensional matching (3DM) problem,
which is known to be NP-hard [14] and formally defined
as follows.
3-Dimensional Matching (3DM)
INSTANCE Set M = {m
1
,m
2
,...,m
m
} ⊆
W × X × Y,where W = {w
1
,w
2
,...,w
q
},
X = {x
1
,x
2
,...,x
q
},and Y = {y
1
,y
2
,...,y
q
}
are disjoint sets having the same number q of
elements and |M| = m.
QUESTION Does M contain a matching,
i.e.,a subset M

= {m

1
,m

2
,...,m

q
} ⊆ M
such that |M

| = q and no two elements of
M

agree in any coordinate?
Given an instance of 3DM,we construct a graph G =
(V,E) as shown in Fig.2,where nodes are distributed
into four layers and edges exist only between nodes in
the same layer or adjacent layers.The graph in Fig.2
is constructed from the following instance of 3DM.
W = {w
1
,w
2
},X = {x
1
,x
2
},Y = {y
1
,y
2
}
and M = {m
1
,m
2
,m
3
,m
4
},where m
1
=
(w
1
,x
2
,y
1
),m
2
= (w
1
,x
1
,y
1
),m
3
= (w
2
,x
2
,y
2
),
and m
4
= (w
2
,x
1
,y
2
).
In the top layer,there are 3 disjoint groups of set
nodes,W = {W
1
,W
2
,...,W
m−q
},X = {X
1
,X
2
,...,X
m−q
},
and Y = {Y
1
,Y
2
,...,Y
m−q
}.In the second layer,there
are 3 corresponding disjoint groups of element nodes,
W = {w
1
,w
2
,...,w
q
},X = {x
1
,x
2
,...,x
q
},and Y =
{y
1
,y
2
,...,y
q
}.W,X,and Y represent W,X,and Y in
the 3DM instance,respectively.W ∪ W forms a clique
of size m,and so do X ∪ X and Y ∪ Y.Besides the
element nodes,the second layer also contains a group
B = {b
1
,b
2
,...,b
m−q
} of bridge nodes.Each bridge
node is adjacent to every set node in the top layer.In
the third layer,there is a group M= {m
1
,m
2
,...,m
m
}
of triplet nodes representing the elements in M.Each
bridge node in the second layer is adjacent to every
triplet node as well.Each triplet node is also adjacent
to the 3 element nodes that occur in the element in M
that it represents.In the bottom layer,there is a group
M

= {m

1
,m

2
,...,m

q
} of matching nodes representing
a potential 3-dimensional matching M

.Each matching
node is adjacent to every triplet node in the third layer.
The transformation is clearly polynomial,and we prove
m

1
m

2
m
1
m
2
m
3
m
4
w
1
w
2
x
1
x
2
y
1
y
2b
1
b
2
W
1
W
2
X
1
X
2
Y
1
Y
2
Figure 1:Reduction from 3DM to CDN.
that M contains a 3-dimensional matching of size q if
and only if G contains m disjoint CDSs.
We start with the “only if” direction.If M contains
a matching of size q,each triplet node in the matching,
its associated element nodes,and a matching node form
a CDS of G.Each of the other m−q CDSs is comprised
of one bridge node,one set node from each of W,X,Y,
and one of the remaining triplet nodes.
We proceed to prove the “if” direction.If G contains
m disjoint CDSs,each CDS must contain exactly one
triplet node because matching nodes are only adjacent
to triplet nodes.
Recall that each one of W ∪ W,X ∪ X,and Y ∪ Y
forms a clique comprised of q element nodes and m−q
set nodes.Since each CDS only contains one triplet
node,it can dominate at most one element node in
each clique via its triplet node.Therefore,in non-trivial
cases where q ≥ 2,each CDS also has to contain at least
one node from each clique as well.On the other hand,
each CDS can have at most one node from each clique
since each clique only has m nodes to be shared by m
CDSs.Clearly,each CDS also contains exactly one node
from each clique.
If a CDS contains a set node,the set node can only be
connected to its triplet node via some bridge node,since
we have proven above that a CDS can not have another
node from the same clique to connect the set node to
its triplet node.Given m−q bridge nodes,it is clear
that at most m−q CDSs can contain a set node.On
the other hand,each CDS can contain at most one set
node fromeach clique,which means at least m−q CDSs
have to contain some set node.Therefore,it must be the
case that there are exactly m−q CDSs each containing
one set node from each clique,while each of the other
q CDSs contains one element node from each clique.
Note that in each of these q CDSs,each element node
has to be directly connected to the triplet node since
there can not be another node from the same clique.
Thus,these q CDSs form a 3-dimensional matching of
size q we need.
3.2 The packet based model
In this section,we analyze network lifetime maxi-
mization problems in the intensively researched packet
based model.In particular,we analyze the complexity
of a number of network lifetime maximization problems
in different routing paradigms,i.e.,unicast,multicast
and broadcast.NP-hardness proofs are presented for
intractable problems,and polynomial time algorithms
are given for tractable problems.
We start with the problemof maximizing the lifetime
of a broadcast session over energy constrained WSNs,
which is formally defined as follows.
Broadcast lifetime
INSTANCE Directed graph G = (V,A).
Specified source s.Positive integer K.
QUESTION Does G have enough power
to broadcast K packets from s to all other
nodes?
The problem of minimum energy broadcast has been
well researched in the literature and proved to be NP-
hard [15,16].However,the complexity of broadcast
lifetime remains open.Flor´een et al.[8] investigated the
problemof maximizing the lifetime of a multicast session
over a network of energy-constrained nodes,where the
network contains some critical nodes that have to be
included in every steiner tree.Therefore,[8] did not
address our problem.
Theorem 2.Broadcast lifetime is NP-hard.
Proof.The NP-hardness of broadcast lifetime can
be proved by slightly adapting the proof of Theorem 1.
In particular,we shall also prove by reducing from the
3-dimensional matching (3DM) problem.
Given an instance of 3DM,we construct a graph G =
(V,E) as shown in Figure 2,where nodes are distributed
into four layers and edges exist only between nodes in
the same layer or adjacent layers.Nodes in each layer
is the same as defined in the proof of Theorem 1.The
only difference is that,in the bottomlayer,there is only
the source node s,which has an energy of m and is ad-
jacent to all the triplet nodes.Each node other than
s has one unit energy.It is clear that the transfor-
mation is polynomial,and we prove that M contains a
3-dimensonal matching of size q if and only if mpackets
can be broadcast.
We start with the “only if” direction.If M contains a
matching of size q,each triplet node in the matching,its
associated element nodes plus the source node s form a
broadcast tree.Each of the other m−q broadcast trees
is composed of one bridge vertex,one set node for each
of W,X,Y,one triplet node,plus the source node s.All
these broadcast trees are node-disjoint (except s) and s
s
m
1
m
2
m
3
m
4
w
1
w
2
x
1
x
2
y
1
y
2b
1
b
2
W
1
W
2
X
1
X
2
Y
1
Y
2
Figure 2:The graph is constructed from the
following 3DM instance.W = {w
1
,w
2
},X =
{x
1
,x
2
},and Y = {y
1
,y
2
}.M = {m
1
,m
2
,m
3
,m
4
},
where m
1
= (w
1
,x
2
,y
1
),m
2
= (w
1
,x
1
,y
1
),m
3
=
(w
2
,x
2
,y
2
),and m
4
= (w
2
,x
1
,y
2
).
has enough power.Thus,one packet can be broadcast
over each of these m broadcast trees,respectively.
We then prove the “if” direction.If m packets can
be broadcast,it is clear that there have to be m node-
disjoint broadcast trees each containing exactly one triplet
node,since s relies on the triplet nodes to forward its
packets.Consequently,in non-trivial cases where q ≥ 2,
each broadcast tree has to contain at least one node
from each clique to broadcast a packet to the nodes in
the cliques.Since each clique has m nodes,each broad-
cast tree has exactly one node from each clique.If a
broadcast tree contains a set node,the set node can
only be connected to the triplet node of that broadcast
tree via a bridge node,since there can not be another
node from the same clique in the broadcast tree.Given
m − q bridge nodes and 3(m − q) set nodes,it must
be the case that there are m− q broadcast tree each
containing one bridge node and one set node from each
clique.Therefore,each of the other q broadcast trees
contains one element node from each clique and the el-
ement nodes have to be adjacent to the triplet node in
the broadcast tree.These q triplet nodes thus form a
3-dimensional matching of size q.
Similarly,in the problem of multicast lifetime,we
want to maximize the number of packets that can be
multicast from a specified source s to a specified group
T of terminals.Since broadcast is just a special case
of multicast,the NP-hardness of multicast lifetime di-
rectly follows.
We then proceed to investigate the problem of maxi-
mizing lifetime of unicast sessions.Similarly,we also de-
fine network lifetime as the maximumnumber of packets
that can be delivered by the network.Because even if
some nodes fail due to battery depletion,the network
may still be able to deliver packets for a unicast session.
There are four different cases in unicast:one-to-one
unicast,one-to-many unicast,many-to-one unicast and
many-to-many unicast,of which many-to-many unicast
is the most general case.Meanwhile,there are two dif-
ferent flow models in unicast,i.e.,the multiple commod-
ity model and the single commodity model.In the multi-
ple commodity model,packets to be delivered between
each source-sink pair are considered a separate com-
modity.In the more relaxed single commodity model
that has been previously studied by Chang and Tassiu-
las [3],all packets are considered the same commodity
and each sink is satisfied if and only if it receives the
number of packets it requests,no matter which source
sends the packets.The most general case of many-to-
many unicast lifetime is formally defined in each model,
respectively.The definitions of the other three cases
can be easily induced as special cases of many-to-many
unicast lifetime.
Many-to-many unicast lifetime (mul-
tiple commodity model)
INSTANCE Directed graph G = (V,A).
Specified set of sources S = {s
1
,s
2
,...,s
m
} ⊆
V and specified set of sinks D = {t
1
,t
2
,...,t
n
} ⊆
V.Each source s
i
has N
ij
packets to be de-
livered to sink t
j
.Positive integer K.
QUESTION Does Ghave enough power to
deliver K packets?
Many-to-many unicast lifetime (single
commodity model)
INSTANCE Directed graph G = (V,A).
Specified set of sources S = {s
1
,s
2
,...,s
m
} ⊆
V and specified set of sinks D = {t
1
,t
2
,...,t
n
} ⊆
V.Each source s
i
has N
s
i
packets to be deliv-
ered and each sink t
j
requests for N
t
j
packets.
Positive integer K.
QUESTION Does Ghave enough power to
deliver K packets?
It is clear that the definition of many-to-one unicast
lifetime,one-to-many unicast lifetime,and one-to-one
unicast lifetime remain the same in the multiple com-
modity model and the single commodity model.
Lemma 1.One-to-one unicast lifetime is NP-hard.
Proof.It suffices to prove the NP-hardness of the
special case where every packet is to be sent from a
source node s to a sink node t.Again,we reduce from
3DM and we illustrate the reduction in Figure 3 with
the same 3DMinstance as used in Figure 2.Nodes are
still distributed into four layers.In the top layer,there
is the sink node t.In the second layer,there are 3 dis-
joint groups of element nodes,W = {w
1
,w
2
,...,w
q
},
s
m
1
m
2
m
3
m
4
w
1
w
2
x
1
x
2
y
1
y
2
t
Figure 3:Reduction from 3DM to one-to-one unicast.
X = {x
1
,x
2
,...,x
q
},and Y = {y
1
,y
2
,...,y
q
}.W,X,
and Y represent W,X,and Y,respectively.Each el-
ement node has an energy of 1 and is adjacent to t.
In the third layer,there is a group of m triplet nodes
M = {m
1
,m
2
,...,m
m
} representing the elements in
M.Each triplet node has an energy of 3 and is adja-
cent to t as well as the three element nodes occurring
in the element in M that it represents.In the bot-
tom layer,there is the source node s with an energy of
m+2q.Each triplet node is also adjacent to s.Edges
between triplet nodes and t have a weight of 3,while
the other edges have a weight of 1.The transformation
is clearly polynomial,and we prove that M contains a
3-dimensional matching of size q if and only if m+2q
packets can be gathered from s to t.
If M contains a 3-dimensional matching of size q,3q
packets can be delivered through the q triplet nodes
in the matching and the 3q element nodes.The other
m−q packets can be delivered through the other m−q
triplet nodes.
If m+2q packets can be delivered from s to t,it is
clear that the only way to achieve that is the same as
described above,since every packet has to be forwarded
by some triplet node.Thus,the q triplet nodes adjacent
to the 3q element nodes form a 3-dimensional matching
of size q.
Since one-to-one unicast lifetime is a special case of
the other three unicast lifetime problems,the following
theorem directly follows.
Theorem 3.In both multiple commodty model and
single commodity model,many-to-many unicast lifetime,
many-to-one unicast lifetime,one-to-many unicast life-
time and one-to-one unicast lifetime are all NP-hard.
Although the unicast lifetime problems are proven to
be NP-hard,it turns out that in cases where each node
i has a fixed transmission power of P
max
(i) (e.g.tiny
sensor nodes may not be able to adjust their transmis-
sion power),we may be able to solve themin polynomial
time.
Theorem 4.If each node has a fixed transmission
power,one-to-one unicast lifetime can be solved in poly-
nomial time.
Proof.Given an instance of one-to-one unicast life-
time,for each node i ∈ V,define its capacity to be
c
i
= p
i
/P
max
(i),where p
i
is its initial energy.An al-
gorithm for the node-capacitated network flow problem
[17] can be applied to compute the maximum number
of packets that can be delivered from s to t.
Theorem 5.If each node has a fixed transmission
power,many-to-one unicast lifetime can be solved in
polynomial time.
Proof.Given that one-to-one unicast lifetime is poly-
nomially solvable,it suffices to reduce many-to-one uni-
cast lifetime to one-to-one unicast lifetime.First of all,
we point out that for many-to-one unicast lifetime,we
can safely assume without loss of generality that t/∈ S.
Given an instance of many-to-one unicast lifetime,we
transform it into an instance of one-to-one unicast life-
time as follows.For each source node s
i
,generate a
mirror node s

i
with an energy of n
i
,where n
i
is the
number of packets to be delivered from s
i
to the sink t.
Then,add a directed link of weight 1 from s

i
to s
i
.Fi-
nally,add a super source s and a directed link of weight
0 froms to each mirror node.All packets are now to be
delivered froms to t.It is clear that the transformation
is polynomial.
Assume that K packets can be delivered to t in the
given instance of many-to-one unicast lifetime,where
each source node s
i
has n

i
≤ n
i
packets delivered to t.
In the constructed instance of one-to-one unicast life-
time,s can safely dispatch the n

i
packets to s
i
via s

i
,
and all the K packets can be delivered to t along the
same paths as they travel along in the given instance of
many-to-one unicast lifetime.
On the other hand,if K packets can be delivered
from s to t in the constructed one-to-one unicast life-
time instance,each packet has to travel through some
source node s
i
.The available energy at mirror nodes
guarantees that for each 1 ≤ i ≤ m,at most n
i
pack-
ets first reaches s
i
among the source nodes.Thus,in
the given instance of many-to-one unicast lifetime,K
packets can travel from the sources to t along the same
paths as they travel along in the constructed instance
of one-to-one unicast lifetime.
Theorem 6.If each node has a fixed transmission
power,one-to-many unicast lifetime can be solved in
polynomial time.
Proof.We similarly prove by reducing to one-to-
one unicast lifetime and point out that for one-to-many
unicast lifetime,we can also safely assume without loss
of generality that s/∈ D.
Given an instance of one-to-many unicast,we trans-
formit into an instance of one-to-one unicast lifetime as
follows.For each sink node t
i
,generate a mirror node t

i
with an energy of n
i
,where n
i
is the number of packets
to be delivered from the source node s to t
i
.And add
a directed link of weight 0 from t
i
to t

i
.Then,add a
super sink t and a directed link of weight 1 from each
mirror node to t.All packets are now destined to t.It
is clear that the transformation is polynomial.
Assume that K packets can be delivered in the given
instance of one-to-many unicast lifetime,where each
sink node t
i
receives n

i
≤ n
i
packets.Then in the con-
structed instance of one-to-one unicast lifetime,each
sink node t
i
can simply forward the n

i
packets to t via
t

i
,and all the K packets are thus delivered to t.
On the other hand,if K packets can be delivered from
s to t in the constructed instance of one-to-one unicast
lifetime,each packet has to travel through some sink
node t
i
.The available energy at mirror nodes guaran-
tees that for each 1 ≤ i ≤ n,at most n
i
packets travel
to t via t

i
.Thus,in the given instance of one-to-many
unicast lifetime,K packets can be delivered along the
same paths as they travel along in the constructed in-
stance of one-to-one unicast lifetime.
Theorem 7.In the single commodity model,if each
node has a fixed transmission power,many-to-many uni-
cast lifetime is polynomially solvable.
Proof.In the single commodity model,packets can
be delivered fromany source to any sink.Thus,we only
need to consider non-trivial cases where S∩D = φ.Re-
call that many-to-one unicast lifetime remains the same
in the multiple commodity model and the single com-
modity model.Given Theorem 5,it suffices to reduce
many-to-many unicast lifetime to many-to-one unicast
lifetime.
Given an instance of many-to-many unicast lifetime,
generate a mirror node t

i
with an energy of n
i
for each
sink t
i
,where n
i
is the number of packets requested by
t
i
.Add a directed link (t
i
,t

i
) of weight 0.Then,add a
super sink t and a directed link of weight 1 from each
mirror node to t.All packets are now destined to t.The
transformation is clearly polynomial.
Assume that K packets can be delivered in the given
instance of many-to-many unicast lifetime,where each
sink node t
i
receives n

i
≤ n
i
packets.Then in the
constructed instance of many-to-one unicast lifetime,
each sink node t
i
can simply forward the n

i
packets to
t via t

i
,and all the K packets are thus delivered to t.
On the other hand,if K packets can be delivered to t
in the constructed instance of many-to-one unicast life-
time,each packet has to travel through some sink node
t
i
.The available energy at mirror nodes guarantees that
for each 1 ≤ i ≤ n,at most n
i
packets travel to t via
t

i
.Thus,in the given instance of many-to-many uni-
cast lifetime,K packets can be delivered along the same
paths as they travel along in the constructed instance
of many-to-one unicast lifetime.
Theorem 8.In the multiple commodity model,even
if each node has a fixed transmission power,many-to-
many unicast lifetime remains NP-hard.
Proof.We prove by reducing fromthe NP-hard dis-
joint connecting paths problem [14],which is defined as
follows.
Disjoint connecting paths
INSTANCE Graph G = (V,E),where V
is the set of nodes and E is the set of edges.
Disjoint set of sources S = {s
1
,s
2
,...,s
m
} ⊆
V and set of sinks D = {t
1
,t
2
,...,t
m
} ⊆ V.
QUESTION Does G contain m node dis-
joint paths,each connecting one pair of source
and sink (s
i
,t
i
) for all 1 ≤ i ≤ m?
Given an instance of disjoint connecting paths,assign
each edge a weight of 1.Assign each sink node an energy
of 0 and each non-sink node an energy of 1.Let each
s
i
have one packet to be delivered to the correspond-
ing sink t
i
.The transformation is clearly polynomial,
and we show that m packets can be delivered from the
sources to the sinks if and only if there are m node dis-
joint paths each connecting one pair of source and sink
(s
i
,t
i
) for all 1 ≤ i ≤ m.
If G contains m node disjoint paths each connecting
one pair of source and sink (s
i
,t
i
) for all 1 ≤ i ≤ m,each
s
i
can deliver its packet to t
i
along the path connecting
them.And all of the m packets can thus be delivered.
Assume that all of the m packets can be delivered.
Since each edge has a weight of 1 and non-sink nodes
have an energy of 1,each non-sink node is on the deliv-
ery path of at most one packet.Sink nodes do not have
energy and there is only one packet destined to each
sink,thus each sink node is on the delivery path of at
most one packet as well.Therefore,the delivery paths
of the m packets are node disjoint,each connecting one
pair of source and sink (s
i
,t
i
) for all 1 ≤ i ≤ m.
4.CONCLUSIONS
We have presented formal analysis of a variety of net-
work lifetime maximization problems in different en-
ergy consumption models.An analysis of energy con-
sumption in wireless sensor networks leads to two en-
ergy consumption models for formal analysis,i.e.,the
time based model and the intensively researched packet
based model.Various network lifetime maximization
problems are identified in individual models.The com-
plexity of these problems are formally analyzed.
Most of the past research efforts aiming to extend
network lifetime are based on the packet based model,
while it is well known that in many applications energy
consumption in the time based model is comparable to
that in the packet based model.On the other hand,
there are two different approaches to network lifetime
maximization,and most of the past research efforts fol-
lowed the indirect approach of energy conservation.Al-
though helpful to extend network lifetime,energy con-
servation is not precisely the same problem as network
lifetime maximization.
In this paper,we directly investigate the problem of
network lifetime maximization in individual energy con-
sumption models as well as routing paradigms.In the
time based model,we study the problem of maximiz-
ing network lifetime while preserving connectivity and
prove that it is NP-hard.In the packet based model,we
formally define the following problems:broadcast life-
time,multicast lifetime,many-to-many unicast lifetime,
many-to-one unicast lifetime,one-to-many unicast life-
time and one-to-one unicast lifetime.Broadcast lifetime
and multicast lifetime are NP-hard,even if each node
has a fixed transmission power.We show that the uni-
cast lifetime problems are NP-hard in both the multi-
ple commodity model and the single commodity model.
However,we show that in cases where each node has
a fixed transmission power,many-to-one unicast life-
time,one-to-many unicast lifetime,and one-to-one uni-
cast lifetime are polynomially solvable.Many-to-many
unicast lifetime is also polynomially solvable in the sin-
gle commodity model,but remains NP-hard in the mul-
tiple commodity model.
5.REFERENCES
[1] C.Ulmer,L.Alkalai,and S.Yalamanchili,
“Wireless distributed sensor networks for in-situ
exploration of mars,” NASA,Tech.Rep.,2003.
[2] J.M.Kahn,R.H.Katz,and K.S.J.Pister,
“Next century challenges:Mobile networking for
“smart dust”,” in ACM MobiCom,August 1999,
pp.271–278.
[3] J.-H.Chang and L.Tassiulas,“Routing for
maximizing system lifetime in wireless ad-hoc
networks,” in Proceedings of 37th Annual Allerton
Conference on Communication,Control,and
Computing,1999.
[4] J.H.Chang and L.Tassiulas,“Energy conserving
routing in wireless ad-hoc networks,” in IEEE
INFOCOM,2000.
[5] A.K.Das,R.J.Marks,M.El-Sharkawi,
P.Arabshahi,and A.Gray,“Mdlt:A polynomial
time optimal algorithm for maximization of
time-to-first-failure in energy constrained wireless
broadcast networks,” in Proceedings of IEEE
GLOBECOM 2003,December 1–5 2003.
[6] I.Kang and R.Poovendran,“Maximizing static
network lifetime of wireless broadcast adhoc
networks,” in IEEE ICC,2003.
[7] A.K.Das,M.El-Sharkawi,R.J.Marks,
P.Arabshahi,and A.Gray,“Maximization of
time-to-first-failure for multicasting in wireless
networks:Optimal solution,” in Proceedings of
MILCOM 2004,Oct.31 – Nov.3 2004.
[8] P.Flor´een,P.Kaski,J.Kohonen,and
P.Orponen,“Multicast time maximization in
energy constrained wireless networks,” in
DIALM-POMC,2003.
[9] M.Stemm and R.H.Katz,“Measuring and
reducing energy consumption of network
interfaces in hand-held devices,” IEICE
Transactions on Communications,vol.E80-B,
no.8,pp.1125–1131,1997.
[10] S.Singh and C.S.Raghavendra,“Pamas:Power
aware multi-access protocol with signalling for ad
hoc networks,” ACM Computer Communication
Review,vol.28,no.3,pp.5–26,July 1998.
[11] O.Kasten,“Energy consumption,” 2001.[Online].
Available:http://www.inf.ethz.ch/kas-
ten/research/bathtub/energy
consumption.html
[12] D.M.Blough and P.Santi,“Investigating upper
bounds on network lifetime extension for
cell-based energy conservation techniques in
stationary ad hoc networks,” in ACM MobiCom,
2002,pp.183–192.
[13] S.Guha and S.Khuller,“Approximation
algorithms for connected dominating sets,” in
Proceedings of the 4th European Symposium on
Algorithms (ESA 1996),1996,pp.179–193.
[14] M.R.Garey and D.S.Johnson,Computers and
Intractability:A Guide to the Theory of
NP-Completeness.W.H.Freeman and Company,
1979.
[15] M.Cagalj,J.Hubaux,and C.Enz,
“Minimum-energy broadcast in all-wireless
networks:Np-completeness and distribution
issues,” in ACM MobiCom,2002.
[16] W.Liang,“Constructing minimum-energy
broadcast trees in wireless ad hoc networks,” in
ACM MobiHoc,June 2002.
[17] T.H.Cormen,C.E.Leiserson,R.L.Rivest,and
C.Stein,Introduction to Algorithms (Second
Edition).MIT Press,2001.